Science of the Total Environment 653 (2019) 938–946

Contents lists available at ScienceDirect

Science of the Total Environment

j ourna l homepage: www.e lsev ie r .com/ locate /sc i totenv

Airborne Alternaria and Cladosporium fungal spores in Europe: Forecastingpossibilities and relationships with meteorological parameters

Agnieszka Grinn-Gofroń a, Jakub Nowosad b,c, Beata Bosiacka a, Irene Camacho d, Catherine Pashley e,Jordina Belmonte f,g, Concepción De Linares f,g, Nicoleta Ianovici h, Jose María Maya Manzano i,Magdalena Sadyś j,k, Carsten Skjøth j, Victoria Rodinkova l, Rafael Tormo-Molina m, Despoina Vokou n,Santiago Fernández-Rodríguez m, Athanasios Damialis n,o,⁎a Department of Plant Taxonomy and Phytogeography, Faculty of Biology, University of Szczecin, Szczecin, Polandb Space Informatics Lab, University of Cincinnati, 219 Braunstein Hall, Cincinnati, OH 45221, USAc Institute of Geoecology and Geoinformation, AdamMickiewicz University, Poznan, Polandd Madeira University, Faculty of Life Sciences, Campus Universitário da Penteada, 9000-390 Funchal, Portugale Institute for Lung Health, Department of Infection, Immunity and Inflammation, University of Leicester, Leicester LE1 7RH, UKf Unidad de Botánica, Facultad de Ciencias, Universidad Autónoma de Barcelona, Barcelona, Spaing Botany Unit, Dept. Of Animal Biology, Plant Biology and Ecology, Universitat Autònoma de Barcelona, Bellaterra, Cerdanyola del Vallès, Spainh West University of Timisoara, Department of Biology, Faculty of Chemistry-Biology-Geography, Romaniai University of Extremadura, Department of Plant Biology, Ecology and Earth Sciences, Faculty of Science, Avda Elvas s/n, 06071 Badajoz, Spainj University of Worcester, Institute of Science and the Environment, Henwick Grove, Worcester WR2 6AJ, United Kingdomk Hereford & Worcester Fire and Rescue Service Headquarters, Performance & Information, Hindlip Park, Worcester, WR3 8SP, United Kingdoml National Pirogov Memorial Medical University, Vinnytsia, Ukrainem Department of Construction, School of Technology, University of Extremadura, Avda. de la Universidad s/n, Cáceres, Spainn Department of Ecology, School of Biology, Aristotle University of Thessaloniki, Thessaloniki GR-54124, Greeceo Chair and Institute of Environmental Medicine, UNIKA-T, Technical University of Munich and Helmholtz Zentrum München, Germany - German Research Center for Environmental Health,Neusaesser Str. 47, DE-86156 Augsburg, Germany

H I G H L I G H T S G R A P H I C A L A B S T R A C T

• No operational forecasting model for al-lergenic fungal spore exposure exists inEurope.

• Potential exposure in Europe wasassessed and predicted for 2major aller-genic fungi.

• Random forestmodellingwas applied toN7000 daily time series.

• Air temperature and vapour pressurewere the most significant variables.

• Classification models showed higher ca-pacity for large-scale spore predictions.

⁎ Corresponding author at: Chair and Institute of EnvirE-mail addresses: agofr@univ.szczecin.pl (A. Grinn-Go

jmmaya@unex.es (J.M.M. Manzano), magdalenasadys@w

https://doi.org/10.1016/j.scitotenv.2018.10.4190048-9697/© 2018 Elsevier B.V. All rights reserved.

a b s t r a c t

a r t i c l e i n f o

Article history:Received 26 June 2018Received in revised form 7 October 2018Accepted 30 October 2018Available online 05 November 2018

Airborne fungal spores are prevalent components of bioaerosols with a large impact on ecology, economy andhealth. Their major socioeconomic effects could be reduced by accurate and timely prediction of airbornespore concentrations. The main aim of this study was to create and evaluate models of Alternaria andCladosporium spore concentrations based on data on a continental scale. Additional goals included assessment

onmental Medicine, UNIKA-T, Technical University of Munich, Neusaesser Str. 47, 86156 Augsburg, Germany.froń), camire@uma.pt (I. Camacho), chp5@leicester.ac.uk (C. Pashley), Jordina.Belmonte@uab.cat (J. Belmonte),p.pl (M. Sadyś), santiferro@unex.es (R. Tormo-Molina), vokou@bio.auth.gr (D. Vokou), thanos.damialis@tum.de (A. Damialis).

939A. Grinn-Gofroń et al. / Science of the Total Environment 653 (2019) 938–946

Editor: Pavlos Kassomenos

of the level of generalization of themodels spatially and description of the mainmeteorological factors influenc-ing fungal spore concentrations.Aerobiological monitoringwas carried out at 18 sites in six countries across Europe over 3 to 21 years dependingon site. Quantile random forest modelling was used to predict spore concentrations. Generalization of theAlternaria and Cladosporium models was tested using (i) one model for all the sites, (ii) models for groups ofsites, and (iii) models for individual sites.The study revealed the possibility of reliable prediction of fungal spore levels using gridded meteorological data.The classification models also showed the capacity for providing larger scale predictions of fungal spore concen-trations. Regressionmodels were distinctly less accurate than classificationmodels due to several factors, includ-ing measurement errors and distinct day-to-day changes of concentrations. Temperature and vapour pressureproved to be the most important variables in the regression and classification models of Alternaria andCladosporium spore concentrations.Accurate and operational daily-scale predictive models of bioaerosol abundances contribute to the assessmentand evaluation of relevant exposure and consequently more timely and efficient management of phytopatho-genic and of human allergic diseases.

© 2018 Elsevier B.V. All rights reserved.

Keywords:Advanced statistical modelsAerobiologyBioaerosolsBiometeorologyContinental scaleMolds

1. Introduction

Fungal spores are one of the most prevalent components ofbioaerosols, found across a wide range of biogeographic regionsover long time periods each year. The primary source of spore emis-sions are the substrates on which fungi grow, such as plants, soil anddecaying organic matter. This means the majority of fungal spores inthe air originate from farms, forest stands and green spaces in gen-eral, and decomposing plant material (Bowers et al., 2013). Theywill remain airborne for variable amounts of time and will betransported over distances ranging from a few centimeters to hun-dreds of kilometers (Ansari et al., 2015; Heald and Spracklen, 2009;Stockmarr et al., 2007).

Alternaria and Cladosporium are ubiquitous asexually reproducingfungal genera that produce spores, known as conidia, which are readilyairborne. Both genera contain plant pathogens (Carlile et al., 2007;Chaerani and Voorrips, 2006; Lee et al., 1997; Nowicki et al., 2012;Thomma et al., 2005), many of which also produce phytotoxic metabo-lites that affect mammalian cells (De Lucca, 2007; Friesen et al., 2008;Mamgain et al., 2013). Fungal spores are also important aeroallergens,causing adverse health effects (Krouse et al., 2002).

Aerobiological surveys reported Alternaria as one of themost preva-lent airborne fungal types and an important aeroallergen (Budd, 1986;Mitakakis et al., 2001). Also, Cladosporium is frequently reported as themost abundant aeroallergen and the second most allergenic fungaltype worldwide (Tariq et al., 1996). Allergy to fungi from both generahas been responsible for hospital admissions due to severe asthma at-tacks in sensitised individuals, particularly among children (Bush andProchnau, 2004; Dales et al., 2000). Due to their ubiquitous nature(Damialis et al., 2017; Jedryczka, 2014), exposure to these aeroallergensis literally inevitable during their dispersion season.

It is important to assess the potential risks of the atmospheric pres-ence of fungal spores, and more research is needed to evaluate this(Beggs, 2004; Crameri et al., 2014). There is limited data on long-termtrends in airborne fungal spore abundance, in part due to a lack of rep-resentative fungal spore data sets longer than a decade across the globe.Airborne fungal spores as allergens have received comparatively less at-tention than pollen, and associated public health consequences arelikely to have been underestimated (Damialis et al., 2015).

Sporulation and dispersion of fungi are influenced by several mete-orological factors, including air temperature, relative humidity, precipi-tation, atmospheric turbulence, wind speed and UVB radiation (Al-Subai, 2002; Carlile et al., 2007; Cecchi et al., 2010; Straatsma et al.,2001). However, the exact influence of climatic variability on fungalecology at a larger scale is still not understood. There are indicationsthat changing climate may lead to alterations in phenology (Cordenet al., 2003; Gange et al., 2007; Kauserud et al., 2010) and dynamics offungal communities (Gange et al., 2011).

This complexity of biological and ecological processes representsone of the key problems of modelling biological systems. There are in-consistencies regardingwhat drives and controls the distribution of fun-gal bioaerosols both at a local and regional scale. Multiplemeteorological factors may alter the spatiotemporal distribution ofAlternaria and Cladosporium spores (Corden et al., 2003; Damialis andGioulekas, 2006; De Linares et al., 2010; Escudero et al., 2011; Iglesiaset al., 2007; Recio et al., 2012; Sindt et al., 2016; Skjøth et al., 2012).The relationships between fungal development and environmental fac-tors, including major climatic variables, are often the only componentused for disease forecasting systems (van Maanen and Xu, 2003).

Atmospheric dispersion models have been used to describe the spa-tiotemporal dispersal of fungal pathogens (Burie et al., 2012; Oteroset al., 2015; Stockmarr et al., 2007; Van Leuken et al., 2016). Addition-ally, descriptive, predictive or conceptual modelling of airborne fungalspores concentration is another promising tool, albeit challenging(Grinn-Gofroń and Bosiacka, 2015; Grinn-Gofroń et al., 2018; Iglesiaset al., 2007; Jedryczka et al., 2015). At a minimum, high-resolutiondata on meteorological and geographic variables should be included insuchmodels. It is important to include and analyse all associated factors,so that the resulting models can accurately describe the complex envi-ronmental inter-dependencies.

Given the widespread, ecological, economical and health impact ofspores from Alternaria and Cladosporium, the main goal of the presentstudy was to build models that generalise beyond the observed dataand are capable of estimating the spatiotemporal distribution and con-centration of fungal spores on a broad scale and to address the factorsthat influence the fungal spatiotemporal patterns. Created modelscould help to answer several questions - (i) is it possible to make a reli-able prediction of fungal spore concentrations using only onemodel forall sites, (ii) is there a difference in quality between onemodel for all thesites, models for the groups of sites, andmodels for individual sites, (iii)what are the main meteorological factors influencing fungal spore con-centrations, and (iv) do these factors differ between created models(analysed sites)?

2. Materials and methods

2.1. Aerobiological data

The taxa selected for this study were those of Alternaria andCladosporium. These comprise some of the most well studied fungal taxaworldwide due to their allergenic and phyopathogenic properties. In thecurrent research, the spore abundance ranges greatly from N15% to over96% of the total annual spore index per site (Table 2). This variationalone challenges for the elaboration of a universal forecasting model inEurope. But even if we had added the rest of the fungal diversity persite, this would have not most probably provided any additional insight

940 A. Grinn-Gofroń et al. / Science of the Total Environment 653 (2019) 938–946

to either ecological/biological processes or health impacts, as responses ofdifferent fungi are highly individualistic and their sensitivity (or not) toenvironmental stress (i.e. climate change) would not be reflected reliably(e.g. Damialis et al., 2015). Also, inmost stations in theworld,mainly thesetwo types alone are counted. So as to obtain similar data in such a large-scale spatial study design and consequently comparable results, we didnot extend to the investigation of additional fungal taxa.

Fungal spores were collected in the frame of long-term aerobiologi-cal monitoring in 18 sites from six countries across Europe and for atime span of 1987–2015 (Fig. 1, Table 1). Alternaria spore concentra-tions were measured at all sites and Cladosporium spore concentrationsat 15 out of the 18 (Table 1).

In brief, for this research, all data providers have been inquired re-gardingmajor changes inmethodological procedures and no significantalteration has been reported. Moreover, in all participating countries,microscopic identification of airborne fungal spores has been conductedby expertswith long-standingexperience in such techniques. Therefore,in all stations, the same method of collection and analysis was used.Standard sampling, processing and analysis techniques were followed,according to the recommendations of the European Aerobiology Society(Frenguelli, 2003) and the British Aerobiology Federation (1995). Ateach location, a Hirst-type volumetric spore trap was used to sampleairborne fungal spores (Hirst, 1952), which is considered the gold-standard device for sampling airborne particles of biological origin(Galán et al., 2014). Samples were collected and analysed weekly, ap-plying standardmethods for sample processing andmicroscopic identi-fication (e.g. British Aerobiology Federation, 1995; Galán et al., 2014;Grant Smith, 1984). Final measurements referred to daily resolutionand were expressed as concentrations of fungal spores per cubicmeter of air on a given date.

2.2. Meteorological data

Eight meteorological factors were included as co-factors in the dataanalysis, namely maximum temperature, minimum temperature,

Fig. 1. Location of pollenmonitoring sites against the background of biogeographical regions inper the needs of the current research).

average temperature, vapour pressure, sum of precipitation, potentialevaporation from a free water surface, potential evapotranspirationfrom a crop canopy, and total global radiation. These were acquiredfrom the AGRI4CAST Interpolated Meteorological Database (Baruthet al., 2007).

Based on maximum (Tmax) and minimum (Tmin) temperature, anadditional ninth parameter was included, growing degree days (GDD).Cumulative GDD is an indicator measuring a heat accumulation, andas a proxy, can represent plant and fungal development.

GDD value is calculated as follows:

GDD ¼ Tmax þ Tmin

2Tbase

Values of GDDs were accumulated starting from January 1. GDD donot accumulate when the daily mean temperature (Tmax + Tmin / 2)is lower or equal to the base temperature. Value of Tbase was set to 5.No relevant previous information on base temperatures for fungalspores existed, so we assumed that these would be similar to later-flowering plant species (being abundant mainly during May–August)and which could be found across a variety of latitudes and climates inEurope; therefore, we set the base temperature at 5, in a similarmannerto studies on grass pollen (Emberlin, 1993; Frenguelli et al., 1989).

All the abovementioned factors were considered either based ontheir availability, or because of the focus of the current study, or basedon previous literature on the topic. For instance, relative humiditydata are not available via AGRI4CAST as these cannot be corrected for al-titude differences among sites within the target climatic grid cell. Re-garding other factors, like wind vectors (speed, persistence anddirection), these were excluded from this analysis, as this would makemore sense in a smaller-scale temporal data processing, when the inter-mittent nature of wind would be possible to take into account. Finally,there have been previously published reports of particular meteorolog-ical factors in site-specific studies, included in the current analysis,proven to be the most decisive for obtaining accurate and reliable

Europe (according to the Europaean Environmental Agency: www.eea.eu.int, modified as

Table 1List of monitoring sites, their exact locations, the years of sampling and the fungal genera examined. Alt: Alternaria, Clad: Cladosporium.

Site Country Longitude Latitude Start End Missing years Genera examined

Barcelona Spain 2.164722 41.39361 1995 2015 Alt & CladBellaterra Spain 2.107778 41.50056 1995 2015 Alt & CladDerby UK −1.496360 52.93862 1987 2006 Alt & CladDon Benito Spain −5.833333 38.96667 2011 2014 AltGirona Spain 2.823056 41.98417 1996 2015 1997, 1998 Alt & CladLeicester UK −1.122710 52.62316 2007 2015 Alt & CladLleida Spain 0.595556 41.62806 1996 2015 1997, 1998 Alt & CladManresa Spain 1.839722 41.72000 1996 2015 1998, 1999 Alt & CladPlasencia Spain −6.086855 40.04897 2011 2014 AltRoquetes-Tortosa Spain 0.493056 40.82028 2006 2015 Alt & CladSzczecin Poland 14.541667 53.44167 2004 2014 Alt & CladTarragona Spain 1.243611 41.12000 1996 2015 1998, 1999 Alt & CladThessaloniki Greece 22.950000 40.61667 1987 2004 2001, 2002 Alt & CladTimisoara Romania 21.231239 45.74722 2008 2010 Alt & CladVielha Spain 0.797222 42.70222 2004 2015 Alt & CladVinnytsia Ukraine 28.472074 49.23473 2009 2014 Alt & CladWorcester UK −2.233333 52.18333 2006 2010 Alt & CladZafra Spain −6.416667 38.41667 2011 2014 Alt

941A. Grinn-Gofroń et al. / Science of the Total Environment 653 (2019) 938–946

forecasts on fungal spore concentrations (e.g. Damialis and Gioulekas,2006).

2.3. Models

Partial autocorrelation function (Durbin, 1960) was applied inde-pendently to Alternaria and Cladosporium daily counts to check for thetemporal autocorrelation of spore data. This summarises the relation-ship between an observation xt and observations with lagged timesteps (days) removing the impact of the values at all shorter lags.

Redundancy among the lagged values of meteorological predictorswas explored using principal coordinates analysis (PCA) (Jolliffe,1986). PCA transforms a number of correlated variables into a smallerset of uncorrelated variables. The role of PCA in this studywas to reducethe dimension of the data, reduce computational time of models build-ing, minimise spurious effects of single lags and ease interpretation ofthe final models.

Two main types of modelling techniques were used, regression andclassification. The Quantile Random Forest method was used to createregression models. This is a generalisation of Random Forests thatinfer the full conditional distribution of a response variable (Li et al.,2011). We decided not to predict mean value, but rather a medianvalue, because of a non-symmetrical distribution of fungal spore values.Daily spore concentrations of Alternaria or Cladosporium were used asdependent variables (model output) in regression models. Breiman'sRandom Forest (Breiman, 2001) was used in classification models. Soas to eliminate high levels of statistical noise, daily spore concentrationswere divided into two levels, low and high, according to the thresholdsof Alternaria and Cladosporium allergens to evoke allergic symptoms(Gravesen, 1979). Alternaria values lower than 100 spores were consid-ered as ‘low’ and beyond that threshold they were characterised as‘high’. For Cladosporium, this threshold was 3000 spores. As airbornefungal spore measurements could still exhibit a huge disparity in thefrequencies of the observed classes, we adopted an optimizing probabil-ity threshold technique (Kuhn and Johnson, 2013; Nowosad, 2016). Inthis approach, alternative cutoffs for the predicted probabilities weredetermined using resampling. Sensitivity (Sens), specificity (Spec), pos-itive predictive value (Ppv), and negative predictive value (Npv) werecalculated for 20 different threshold values. For each model, optimalthreshold value was established minimizing the distance betweenSens, Spec, Ppv, Npv and the value of 1. This value indicates the best pos-sible performance.

Attempting to also reduce the spatial variability among monitoringsites across Europe, analysed sites were divided into three groupsbased on the annual temporal changes of fungal spore concentrationsand on data availability (Table 1), as follows: northeastern Spain

(Barcelona, Bellaterra, Girona, Lleida,Manresa, Roquetes-Tortosa, Tarra-gona, Vielha), western Spain (Don Benito, Plasencia, Zafra) and non-Spanish sites (Derby, Leicester, Szczecin, Thessaloniki, TimisoaraVinnytsia, Worcester). The grouping was based on the seasonality andmulti-modality of airborne spore concentrations, i.e. the first group ex-hibited the highest seasonality and normality of data, with longer sea-sons and fewer outliers. The third group presented a bi- or multi-modal yearly pattern, whilst the second group included those siteswith only Alternaria spore measurements.

The modelling was performed in three ways: 1) per site, 2) pergroup of sites as mentioned above, and 3) an integrated model for allsites combined. This resulted in 22 combinations for Alternaria (18sites, 3 groups of sites, and 1 whole dataset) and 18 combinations forCladosporium (15 sites, 2 groups of sites, and 1 whole dataset) generat-ing a total of 40 regression and 40 classification models.

2.4. Validation metrics

The accuracy of all models was assessed using a repeated k-foldcross-validation (Kuhn and Johnson, 2013). Regression models wereevaluated using the Symmetric Mean Absolute Percentage Error(SMAPE). Thus, accuracy measure was based on relative errors insteadof standard metrics like Root Mean Square Error (RMSE) orMean Abso-lute Error (MAE), which was particularly important in our study be-cause of high among-site data variability. Classification modellingresultswere characterized using balanced accuracy. Thismetric is calcu-lated as sensitivity + specifity/2. It is a better measurement in the caseof imbalanced datasets as it gives the same weight for correctly pre-dicted cases with low concentration and correctly predicted caseswith high concentration. Influence of the predictors was determinedusing a scaled permutation importance (mean decrease in accuracy)(Breiman, 2001).

3. Results

3.1. Spatiotemporal variability of spore concentrations

Fungal spore concentrations differed between Alternaria andCladosporium, between sites, and between years of measurements(Table 2, Figs. 2, 3). The annual sum of Alternaria daily concentrationsvaried between 907 (Vielha) and 67,166 (Lleida) and had a meanvalue of 13,448. In addition to Lleida having the highest average annualsums of Alternaria, the values here were also themost variable betweenyears. Other sites with substantial changes between years were Derby,Thessaloniki, Vielha, andGirona. The annual sumof the daily concentra-tions of Cladosporium varied between 24,637 (Thessaloniki) and

Table 2Summary of fungal spore concentration data per site. Minimum, maximum and average yearly spore concentration is given for the examined genera and their contribution % to the totalannual fungal spore index.

Site Alternaria Cladosporium

Min. Max. Mean % of all fungal taxa Min. Max. Mean % of all fungal taxa

Barcelona 4967 12,925 9310 1.8 84,636 339,962 204,770 38.9Bellaterra 6272 16,430 11,622 1.1 152,177 478,794 311,729 29.4Derby 6805 52,007 24,918 0.7 509,554 1,500,699 950,719 25.3Don Benito 8789 14,878 11,774 25.2Girona 5032 17,696 12,873 0.7 165,634 539,613 361,858 19.9Leicester 16,522 31,831 21,519 0.7 523,894 1,088,832 852,812 26.4Lleida 3282 67,166 35,380 2.0 48,636 1,194,049 579,855 32.1Manresa 7571 37,859 20,981 1.6 144,782 660,514 473,642 35.6Plasencia 2701 5413 3610 10.6Roquetes-Tortosa 10,086 18,080 13,983 0.9 214,236 445,494 323,300 20.0Szczecin 10,333 31,694 17,553 12.3 276,520 813,364 498,877 72.0Tarragona 5580 13,986 10,149 1.1 73,741 323,176 221,793 23.1Thessaloniki 2962 18,925 10,383 11.2 24,637 138,678 74,279 70.5Timisoara 4714 5944 5176 4.5 68,372 175,750 121,541 92.2Vielha 907 3956 1979 0.2 63,672 272,798 162,606 16.4Vinnytsia 10,712 23,976 17,556 4.5 128,951 494,932 279,873 66.9Worcester 6022 9279 7776 0.7 437,171 902,136 660,895 62.5Zafra 3535 9383 5516 14.7All sites 907 67,166 13,448 24,637 1,500,699 405,237

942 A. Grinn-Gofroń et al. / Science of the Total Environment 653 (2019) 938–946

1,500,699 (Derby) and had a mean value of 405,237. Annual spore con-centrations in Lleida were also the most changeable for Cladosporium.Overall, the order of average values for all sites was similar for Alternariaand Cladosporiumwith two exceptions,Worcester and Thessaloniki. Theannual sum of the daily concentrations of Alternaria in Worcester wasrelatively small and invariable, whilst Cladosporium values were highand changeable. In Thessaloniki Alternaria levels were moderate andvaried substantially, whilst Cladosporium annual values were smalland more consistent.

For Alternaria, sites could be split into two groups based on the timecourse of the season: (i) siteswith period of high concentrations and pe-riod of lower concentrations (such as Barcelona, Lleida, etc.), (ii) siteswith one period of high concentrations and period with absence ofAlternaria spores (such as Szczecin, Thessaloniki and Vielha) (Fig. 2).Time course of Cladosporium spore concentrations is more heteroge-neous, with probably two main groups: (i) Spanish sites with periodof high concentration and period of moderate concentrations, (ii) siteswith one period of high concentration and period with low

Fig. 2. Daily concentrations of airborne Alternaria spores (log(1 + x)) per sampling year for allusing cubic spline smoothing technique. Colouration of actual data signifies different time spanreferences to colour in this figure legend, the reader is referred to the web version of this artic

concentrations or absence of spore concentrations (such as Szczecin,Derby, Vinnytsia). Additionally, a third group, consisting of Don Benito,Plasencia, and Zafra, was separated due to missing data of Cladosporium(Fig. 3).

3.2. Predictor variables

Based on the spatiotemporal analysis, lagged daily values of ninemeteorological parameters between 1 and15dayswere created. A prin-cipal component analysis (PCA) for each of the parameters was run andthe results of the PCA gave an insight into the variability of predictors. Inmost of the cases, the two first components explained the majority ofvariations. The cumulative value was between 0.85 for radiation and0.94 for average temperature in these variables. The first componentexpressed the value of a given parameter (large values of loadings forall of the lags), whilst the second parameter expressed the temporalchanges of a given parameter (the largest, positive value for the firstlag, and the lowest, negative value for the last lag). In the case of the

monitoring sites. Coloured lines represent actual data, whereas black ones are fitting liness, with darker colours indicating lack of more recent spore data. (For interpretation of thele.)

Fig. 3.Daily concentrations of airborneCladosporium spores (log(1+x)) per sampling year for allmonitoring sites. Coloured lines represent actual data,whereas black ones arefitting linesusing cubic spline smoothing technique. Colouration of actual data signifies different time spans, with darker colours indicating lack of more recent spore data. (For interpretation of thereferences to colour in this figure legend, the reader is referred to the web version of this article.)

Table 3The final list of predictor variables.

Predictor name Description

GDD_PC1 First principal coordinate of cumulated GDDPRECIPITATION_lag1 Average daily precipitation lagged by one dayPRECIPITATION_lag2 Average daily precipitation lagged by two daysPRECIPITATION_lag3 Average daily precipitation lagged by three daysPRECIPITATION_lag4 Average daily precipitation lagged by four daysE0_PC1 First principal coordinate of potential evaporation

from a free water surfaceE0_PC2 Second principal coordinate of potential evaporation

from a free water surfaceET0_PC1 First principal coordinate of potential

evapotranspiration from a crop canopyET0_PC2 Second principal coordinate of potential

evapotranspiration from a crop canopyRADIATION_PC1 First principal coordinate of average total global

radiationRADIATION_PC2 Second principal coordinate of average total global

radiationTEMPERATURE_AVG_PC1 First principal coordinate of average temperatureTEMPERATURE_AVG_PC2 Second principal coordinate of average temperatureTEMPERATURE_MAX_PC1 First principal coordinate of average maximum

temperatureTEMPERATURE_MAX_PC2 Second principal coordinate of average maximum

temperatureTEMPERATURE_MIN_PC1 First principal coordinate of average minimum

temperatureTEMPERATURE_MIN_PC2 Second principal coordinate of average minimum

temperatureVAPOURPRESSURE_PC1 First principal coordinate of average vapour pressureVAPOURPRESSURE_PC2 Second principal coordinate of average vapour

pressure

943A. Grinn-Gofroń et al. / Science of the Total Environment 653 (2019) 938–946

accumulated GDD, we used only the first component, which explained0.95 of variation. Precipitation values were the most changeable andthe first two components had a cumulative variance of 0.22. Therefore,based on the autocorrelation plot, we used raw values of precipitationfor lags between 1 and 4 days.

The final group of predictors consisted of 19 variables - thefirst prin-cipal component of accumulated GDD, one to four day lags of precipita-tion, and the first two principal components of the rest of themeteorological parameters (Table 3).

3.3. Regression models

3.3.1. Performance of the modelsThe final 40 regression models (22 for Alternaria and 18 for

Cladosporium) were built and results compared using SymmetricMean Absolute Percentage Error (SMAPE) (Table 4). These valuesranged between 0.56 and 0.90 for Alternaria (average value of 0.69),and 0.53 and 0.73 for Cladosporium (average value of 0.61). For allsites, the models gave a value of 0.76 for Alternaria and a value of 0.73for Cladosporium. Those values were slightly higher than values for sep-arate groups and distinctly higher than values for most of the individualsites. Similarly, most of themodels for the site groups performed worsethan the models for the individual sites. Alternaria model had a SMAPEvalue of 0.73 for the first group (average for the models for the individ-ual sites is this group was 0.67), 0.75 for the second group (average of0.74), and 0.69 for the third group (average of 0.61). These differenceswere higher in Cladosporium models with SMAPE of 0.67 for the firstgroup (average of 0.60) and 0.71 for the second group (average of0.59). Only in four Alternaria models, Derby, Worcester, and Vielha,values of SMAPE were higher than values for models of groups of sites.

3.3.2. Variable importanceThe same set of variables seemed to influence the models of both

taxa (Fig. 4). The most important predictors were the first principal co-ordinates of vapour pressure and temperature (minimum, average,maximumand accumulatedGDD). Theywere followed by the first prin-cipal coordinates of evapotranspiration, evaporation and radiation. Thesecondprincipal coordinates of those parameters had small importance.The smallest values were observed for lagged values of precipitation.

3.4. Classification models

3.4.1. Performance of the modelsThe final classification models were compared using the balanced

accuracy metric (Table 4). Average value of balanced accuracy was0.78 for Alternaria and 0.73 for Cladosporium. These values variedamong sites between 0.50 and 0.90 for Alternaria and between 0.50and 0.99 for Cladosporium. Models for all the sites gave similar resultsof 0.80 and 0.78. The Alternaria model gave a balanced accuracy value

Table 4Performance of the regression and classification models. Regression models are describedby the SMAPE (less is better), classificationmodels are described by the balanced accuracy(more is better).

Group Regression models Classification models

Alternaria Cladosporium Alternaria Cladosporium

Barcelona 0.72 0.59 0.72 0.50Bellaterra 0.65 0.57 0.76 0.55Girona 0.62 0.53 0.81 0.80Lleida 0.63 0.63 0.84 0.75Manresa 0.68 0.66 0.78 0.76Roquetes-Tortosa 0.61 0.56 0.78 0.61Vielha 0.90 0.63 0.50 0.50FIRST GROUP 0.73 0.67 0.77 0.68Derby 0.82 0.55 0.89 0.86Leicester 0.73 0.54 0.90 0.81Szczecin 0.69 0.60 0.88 0.83Vinnytsia 0.60 0.56 0.85 0.75Thessaloniki 0.65 0.69 0.80 0.99Worcester 0.75 0.53 0.84 0.81Timisoara 0.66 0.61 0.66 0.75SECOND GROUP 0.75 0.71 0.85 0.83Don Benito 0.56 0.88Plasencia 0.60 0.50Zafra 0.66 0.77THIRD GROUP 0.69 0.80ALL 0.76 0.73 0.80 0.78

944 A. Grinn-Gofroń et al. / Science of the Total Environment 653 (2019) 938–946

of 0.77 for the first group of sites compared to the average of individualsite's models of 0.75, a value of 0.85 for the second group of sites com-paring to the average of individual site's models of 0.81, and a value of0.80 for the third group of sites compared to the average of individual

Fig. 4. Significance of meteorological factors for each input variable for Alternaria and Cladospofactors are sorted in descending order of the mean value of variable significance per fungal tax

site's models of 0.71. Values of a balanced accuracy for the models ofgroups of sites were also higher for Cladosporium. The first group hada value of 0.68 (average for individual sites was 0.66), and the secondgroup had a value of 0.85 (average for individual sites was 0.80).About 44% (15 of 34) of models for individual sites gave worse valuesof balanced accuracy than models for groups of sites.

3.4.2. Variable importanceValues of variable importance for classification models were more

diverse than for regressionmodels. Alternaria and Cladosporium classifi-cation models were influenced mostly by the same predictors (Fig. 4).Temperature (accumulated GDD) was the most important variable,followed by the first and second principal coordinate of vapour pres-sure. The rest of the predictors showed moderate to low importance.Similarly to regression models, predictors with values of precipitationhad the smallest importance.

4. Discussion

The present study revealed that wide-scale, accurate, operationalmodelling of fungal spore abundances is feasible with one universalmodel, which answers the first research question. Of course, there arerestrictions because of annual and spatial variability, which result invarying performance of the obtained models. Based on the modelused, these individualistic responses (also based on the fungal taxon ex-amined) can be decreased to an extent.We found that specificmeteoro-logical factors significantly contributed to the forecasting power, withair temperature playing the leading role. Consequently, the current re-search highlights the possibility and need for developing universal

rium spores - regression models (top) and classification models (bottom). Meteorologicalon.

945A. Grinn-Gofroń et al. / Science of the Total Environment 653 (2019) 938–946

predictive models of airborne fungal spore abundances, something cur-rently lacking and therefore making this study novel.

In the current research, an integrative approach of the variables af-fecting the distribution of these bioaerosols was adopted, on a local ora regional scale, providing a deeper comprehension of the dynamics ofthese fungal taxa. The survey revealed distinct relationships betweenspore concentrations and sites and years of sampling. The overallspore frequency and the annual sum of the daily concentrations ofCladosporiumwere higher than Alternaria's, a finding also reported pre-viously in different regions of the world. In Cartagena (Spain),Cladosporium represented 62.2% of the total spore count and Alternariaonly 5.3%, however, Alternaria was still the second most abundant fun-gal type (Elvira-Rendueles et al., 2013). Likewise, in Bursa (Turkey),Cladosporium represented 88.1% of the total spore count followed byAlternaria at 4.9% (Ataygul et al., 2007). A similar trend was found inThessaloniki (Greece) by Gioulekas et al. (2004) and in Madrid(Spain) by Sabariego et al. (2007).

Two types of models, regression and classification, were built as apart of this study. The goal of the first one was to predict values of thefungal spore concentration in the studied sites, whilst the classificationmodels were created to predict high levels of the fungal spore concen-trations. Classification models were more accurate than regressionmodels. This is due to a number of factors. Concentration values ofAlternaria and Cladosporium were classified into two groups prior tothe modelling. This procedure generalises the fungal spore values and,therefore, the changes in values. As a result, it reveals more generaltrends instead of just showing the local trends and day-to-day differ-ences.Moreover, the obtained values of concentration are an estimationof thewhole population and are prone to random and systematic (bias)errors (Comtois et al., 1999; Oteros et al., 2013). The benefit of using aclassification method is that it reduces the influence of methodologicaldifferences between sites, such as the relative position of samplersand the heights of buildings, or instrumental and human errors.

The life cycles of many fungal pathogens are strongly determined byweather. The airborne spore concentrations are affected by biologicalfactors (reproduction and survival), weather parameters, land use, re-source availability and competition (Boddy et al., 2014). Dispersal andcirculation are highly influenced by wind and rainfall, whilst germina-tion and infection rates are often dependent upon liquid water on theplant surface (sometimes high relative humidity) and species-specificoptimal temperature ranges. In this study the most important meteoro-logical variables were the same for both Alternaria and Cladosporium.This shows that despite the differences in the values of spore concentra-tion, both taxa are mostly affected by the same meteorological factors,such as temperature (minimum, average, maximum and accumulatedGDD) and vapour pressure, which determines the water content inthe air (related to air humidity).

In many aerobiological studies, temperature and relative humidityhave been proven to be the meteorological parameters most signifi-cantly influencing concentrations of Cladosporium and Alternaria spores,with temperature being positively associated and relative humiditynegatively associated (e.g. Grinn-Gofroń and Strzelczak, 2009;O'Connor et al., 2014; Sadyś et al., 2016; Ianovici, 2016; Almeida et al.,2018).

Regarding other variables, global radiation includes both the directsolar radiation and the diffuse radiation resulting from reflected orscattered sunlight, and can be considered as a function of temperature(Meza andVaras, 2000). One of the solar radiation components, UVB ra-diation, is reported to affect the survival of airborne fungal spores duringmovement over long distances through the atmosphere (Al-Subai,2002). This predictor showed moderate importance in our models.

Rain has been often cited as being one of themost influential factorsin reducing airborne pollen (e.g. Damialis et al., 2005). However, precip-itation was identified as the least important factor in our models. Thereare twomain reasons behind the small effect of precipitation on the fun-gal spore models. Firstly, variation of daily precipitation values does not

closely correlate with the values and levels of Alternaria andCladosporium concentrations, both of which are regarded as dryweather spores. For example, it is possible to have a rainless day and ahigh fungal spore concentration (the middle of the season) and tohave a rainless day without airborne fungal spores (off-season). Sec-ondly, precipitation is characterized by high variability, which couldnot be captured in a daily timescale (sumof precipitation). To better un-derstand the effects of rainfall on fungal spore diversity and abundance,and on circulation patterns, finer resolution data (hourly scale), withdifferent statistical techniques (i.e. artificial intelligence models)would be required.

5. Conclusions

• Classification models were more accurate than regressions forAlternaria and Cladosporium fungal spores.

• Regressionmodels gave better results for individual sites compared togrouped sites, resulting potentially from strong effects from local me-teorological conditions.

• Classification models gave better results in grouped sites rather thanfor individual sites, thus, displaying the capacity for accurately provid-ing larger scale predictions of fungal spore concentrations (comparedto the more localised regression models).

• Temperature (in the form of minimal, average, maximum tempera-ture, and accumulated GDD) and vapour pressure were the most im-portant variables in models of Alternaria and Cladosporium, whilstradiation and daily sum of precipitation had a smaller impact on themodels.

Acknowledgements

This research was partly implemented in the framework of the EU-COST Action DiMoPEx (Diagnosis, Monitoring and Prevention ofExposure-Related Noncommunicable Diseases), under Grant NumberCA15129 (EU Framework Programme Horizon 2020).

References

Almeida, E., Caeiro, E., Todo-Bom, A., Ferro, R., Dionísio, A., Duarte, A., Gazarini, L., 2018.The influence of meteorological parameters on Alternaria and Cladosporium fungalspore concentrations in Beja (Southern Portugal): preliminary results. Aerobiologia34, 219–226.

Al-Subai, A.A.T., 2002. Air-borne fungi at Doha, Qatar. Aerobiologia 18, 175–183.Ansari, T.U., Valsan, A.E., Ojha, N., Ravikrishna, R., Narasimhan, B., Gunthe, S.S., 2015.

Model simulations of fungal spore distribution over the Indian region. Atmos. Envi-ron. 122, 552–560.

Ataygul, E., Celenk, S., Canitez, Y., Bicakci, A., Malyer, H., Sapan, N., 2007. Allergenic fungalspore concentrations in the atmosphere of Bursa, Turkey. J. Biol. Environ. Sci. 1,73–79.

Baruth, B., Genovese, G., Leo, O., 2007. CGMS Version 9.2 - User Manual and TechnicalDocumentation. Office for Official Publications of the European Communities,Luxembourg.

Beggs, P.J., 2004. Impacts of climate change on aeroallergens. Past and future. Clin. Exp. Al-lergy 34 (1507–1073).

Boddy, L., Büntgen, U., Egli, S., Gange, A.C., Heegaad, E., Kirk, P.M., et al., 2014. Climate var-iation effects on fungal fruiting. Fungal Ecol. 10, 20–33.

Bowers, R.M., Clements, N., Emerson, J.B.E., Wiedinmyer, C., Hannigan, M.P., Fierer, N.,2013. Seasonal variability in bacterial and fungal diversity of the near-surface atmo-sphere. Environ. Sci. Technol. 47, 12097–12106.

Breiman, L., 2001. Random Forests. University of California https://www.stat.berkeley.edu/~breiman/randomforest2001.pdf, Accessed date: 30 May 2018.

British Aerobiology Federation, 1995. Airborne Pollens and Spores. A Guide to Trappingand Counting. National Pollen and Hayfever Bureau, Rotherham, United Kingdom.

Budd, T.W., 1986. Allergens of Alternaria. Grana 25, 147–154.Burie, J.B., Calonnec, A., Langlais, M., Mammeri, Y., 2012. Modelling the spread of a path-

ogen over a spatially heterogeneous growing crop. Paper Presented at the 2012 IEEE4th International Symposium on Plant Growth Modeling, Simulation, Visualizationand Applications (PMA), Shanghai, China.

Bush, R.K., Prochnau, J.J., 2004. Alternaria-induced asthma. J. Allergy Clin. Immunol. 113,227–234.

Carlile, M.J., Watkinson, S.C., Gooday, G.W., 2007. The Fungi. 2nd edition. Elsevier, UK, TheNetherlands, USA.

946 A. Grinn-Gofroń et al. / Science of the Total Environment 653 (2019) 938–946

Cecchi, L., D'Amato, G., Ayres, J.G., Galan, C., Forastiere, F., Forsberg, B., Gerritsen, J., Nunes,C., Behrendt, H., Akdis, C., Dahl, R., Annesi-Maesano, I., 2010. Projections of the effectsof climate change on allergic asthma: the contribution of aerobiology. Allergy 65,1073–1081.

Chaerani, R., Voorrips, R.E., 2006. Tomato early blight (Alternaria solani): the pathogen,genetics, and breeding for resistance. J. Gen. Plant Pathol. 72, 335–347.

Comtois, P., Alcazar, P., Néron, D., 1999. Pollen counts statistics and its relevance to preci-sion. Aerobiologia 15, 19–28.

Corden, J.M., Millington, W.M., Mullins, J., 2003. Long-term trends and regional variationin the aeroallergen Alternaria in Cardiff and Derby UK – are differences in climate andcereal production having an effect? Aerobiologia 19, 191–199.

Crameri, R., Garbani, M., Rhyner, C., Huitema, C., 2014. Fungi: the neglected allergenicsources. Allergy 69, 176–185.

Dales, R.E., Cakmak, S., Burnett, R.T., Judek, S., Coates, F., Brook, J.R., 2000. Influence of am-bient fungal spores on emergency visits for asthma to a regional children's hospital.Am. J. Respir. Crit. Care Med. 162, 2087–2090.

Damialis, A., Gioulekas, D., 2006. Airborne allergenic fungal spores and meteorologicalfactors in Greece: Forecasting possibilities. Grana 45, 122–129.

Damialis, A., Gioulekas, D., Lazopoulou, C., Balafoutis, C., Vokou, D., 2005. Transport of air-borne pollen into the city of Thessaloniki: the effects of wind direction, speed andpersistence. Int. J. Biometeorol. 49, 139–145.

Damialis, A., Mohammad, A.B., Halley, J.M., Gange, A.C., 2015. Fungi in a changing world:growth rates may be elevated, but spore production will decrease in future climates.Int. J. Biometeorol. 59, 1157–1167.

Damialis, A., Kaimakamis, E., Konoglou, M., Akritidis, I., Traidl-Hoffmann, C., Gioulekas, D.,2017. Estimating the abundance of airborne pollen and fungal spores at variable ele-vations using an aircraft: how high can they fly? Sci. Rep. 7, 44535.

De Linares, C., Belmonte, J., Canela, M., Díaz De La Guardia, C., Alba-Sanchez, F., Sabariego,S., et al., 2010. Dispersal patterns of Alternaria conidia in Spain. Agric. For. Meteorol.150, 1491–1500.

De Lucca, A.J., 2007. Harmful fungi in both agriculture and medicine. Rev. Iberoam. Micol.24, 3–13.

Durbin, J., 1960. The fitting of time-series models. Revue de l'Institut International deStatistique/Rev. Int. Stat. Inst. 28, 233–244.

Elvira-Rendueles, B., Moreno, J., Garcia-Sanchez, A., Vergara, N., Martinez-Garcia, M., et al.,2013. Air-spore in Cartagena, Spain: viable and non-viable sampling methods. Ann.Agric. Environ. Med. 20, 664–671.

Emberlin, J., 1993. Variations in pollen concentrations – the need for accurate monitoring.Clin. Exp. Allergy 23, 45.

Escudero, O., Seijo, M.C., Fernández-González, M., Iglesias, I., 2011. Effects of meteorolog-ical factors on the levels of Alternaria spores on a potato crop. Int. J. Biometeorol. 55,243–252.

Frenguelli, G., 2003. Basic microscopy, calculating the field of view, scanning of slides,sources of error. Postępy Dermatologii i Alergologii 2, 227–229.

Frenguelli, G., Bricchi, E., Romano, B., Mincigrucci, G., Spieksma, F.T.M., 1989. A predictivestudy on the beginning of the pollen season for Gramineae and Olea europaea L.Aerobiologia 5, 64–70.

Friesen, T.L., Faris, J.D., Solomon, P.S., Oliver, R.P., 2008. Host-specific toxins: effectors ofnecrotrophic pathogenicity. Cell Microbiol. 10, 1421–1428.

Galán, C., Smith, M., Thibaudon, M., Frenguelli, G., Oteros, J., Gehrig, R., Berger, U., Clot, B.,Brandao, R., 2014. Pollen monitoring: minimum requirements and reproducibility ofanalysis. Aerobiologia 30, 385–395.

Gange, A.C., Gange, E.G., Sparks, T.H., Boddy, L., 2007. Rapid and recent changes in fungalfruiting patterns. Science 316, 71.

Gange, A.C., Gange, E.G., Mohammad, A.B., Boddy, L., 2011. Host shifts in fungi caused byclimate change? Fungal Ecol. 4, 154–190.

Gioulekas, D., Damialis, A., Mpalafoutis, C., Papakosta, D., Giouleka, P., Patakas, D., 2004.Allergenic fungal spore records (15 years) and relationship with meteorological pa-rameters in Thessaloniki, Greece. J Investig Allergol Clin Immunol 16, 52–59.

Grant Smith, E., 1984. Sampling and identifying allergenic pollens and molds. An Illus-trated Manual for Physicians and Lab Technicians. Blewstone Press, San Antonio,Texas.

Gravesen, S., 1979. Fungi as a cause of allergic disease. Allergy 34, 135–154.Grinn-Gofroń, A., Bosiacka, B., 2015. Effects of meteorological factors on the composition

of selected fungal spores in the air. Aerobiologia 31, 63–72.Grinn-Gofroń, A., Strzelczak, A., 2009. Hourly predictive ANN and MRT models of

Alternaria and Cladosporium spore concentrations in Szczecin (Poland). Int.J. Biometeorol. 53, 555–562.

Grinn-Gofroń, A., Bosiacka, B., Bednarz, A., Wolski, T., 2018. A comparative study of hourlyand daily relationships between selected meteorological parameters and airbornefungal spore composition. Aerobiologia 34, 45–54.

Heald, C.L., Spracklen, D.V., 2009. Atmospheric budget of primary biological aerosol parti-cles from fungal spores. Geophys. Res. Lett. 36, L09806.

Hirst, J., 1952. An automatic volumetric spore trap. Ann. Appl. Biol. 39, 257–265.

Ianovici, N., 2016. Atmospheric concentrations of selected allergenic fungal spores in re-lation to some meteorological factors, in Timisoara (Romania). Aerobiologia 32,139–156.

Iglesias, I., Rodríguez-Rajo, F.J., Méndez, J., 2007. Evaluation of the different Alternaria pre-diction models on a potato crop in A Limia (NW of Spain). Aerobiologia 23, 27–34.

Jedryczka, M., 2014. Aeromycology: studies of fungi in aeroplankton. Folia Biol. et Oecol.10, 18–26.

Jedryczka, M., Strzelczak, A., Grinn-Gofron, A., Nowak, M., Wolski, T., Siwulski, M., et al.,2015. Advanced statistical models commonly applied in aerobiology cannot accu-rately predict the exposure of people to Ganoderma spore-related allergies. Agric.For. Meteorol. 201, 209–217.

Jolliffe, I.T., 1986. Principal Component Analysis. Second edition. Springer, pp. 1–518.Kauserud, H., Heegaard, E., Semenov, M.A., Boddy, L., Halvorsen, R., Stige, L.C., Sparks, T.H.,

Gange, A.C., Stenseth, N.C., 2010. Climate change and spring-fruiting fungi. Proc. RoyalSoc. Biol. Sci. 277, 1169–1177.

Krouse, J.H., Chadwick, S.J., Gordon, B.R., Derebery, M.J., 2002. Allergy and immunology.An Otolaryngic Approach. Lippincott Williams & Wilkins, USA.

Kuhn, M., Johnson, K., 2013. Applied Predictive Modeling. Springer, New York, pp. 1–595.Lee, K., Young, K., Kang, H., Ahn, K., Min, K., Cha, B., 1997. Cucumber scab caused by

Cladosporium cucumerinum in Korea. Korean J. Plant Pathol. 13, 288–294.Li, R., Peng, L., Road, C., 2011. Quantile regression for left-truncated semicompeting risks

data. Biometrics 7, 701–710.Mamgain, A., Roychowdhury, R., Tah, J., 2013. Alternaria pathogenicity and its strategic

controls. Res. J. Biol. 1, 1–9.Meza, F., Varas, E., 2000. Estimation of mean monthly solar global radiation as a function

of temperature. Agric. For. Meteorol. 100, 231–241.Mitakakis, T.Z., Barnes, Ch., Tovey, R., 2001. Spore germination increases allergen release

from Alternaria. J. Allergy Clin. Immunol. 107, 338–381.Nowicki, M., Nowakowska, M., Niezgoda, A., Kozik, E., 2012. Alternaria black spot of cruci-

fers: symptoms, importance of disease, and perspectives of resistance breeding.Veget. Crops Res. Bull. 76, 5–19.

Nowosad, J., 2016. Spatiotemporal models for predicting high pollen concentration levelof Corylus, Alnus, and Betula. Int. J. Biometeorol. 60, 843–855.

O'Connor, D.J., Sadyś, M., Skjøth, C.A., Healy, D.A., Kennedy, R., Sodeau, J.R., 2014. Atmo-spheric concentrations of Alternaria, Cladosporium, Ganoderma and Didymella sporesmonitored in Cork (Ireland) and Worcester (England) during the summer of 2010.Aerobiologia 30, 397–411.

Oteros, J., Galán, C., Alcázar, P., Domínguez-Vilches, E., 2013. Quality control in bio-monitoring networks, Spanish Aerobiology Network. Sci. Total Environ. 443,559–565.

Oteros, J., García-Mozo, H., Alcázar, P., Belmonte, J., Bermejo, D., Boi, M., et al., 2015. A newmethod for determining the sources of airborne particles. J. Environ. Manag. 155,212–218.

Recio, M., del Mar Trigo, M., Docampo, S., Melgar, M., García-Sánchez, J., Bootello, L., et al.,2012. Analysis of the predicting variables for daily and weekly fluctuations of two air-borne fungal spores: Alternaria and Cladosporium. Int. J. Biometeorol. 56, 983–991.

Sabariego, S., Díez, A., Gutiérrez, M., 2007. Monitoring of airborne fungi inMadrid (Spain).Acta Bot. Croat. 66, 117–126.

Sadyś, M., Adams-Groom, B., Herbert, R.J., Kennedy, R., 2016. Comparisons of fungal sporedistributions using air sampling atWorcester, England (2006–2010). Aerobiologia 32,619–634.

Sindt, C., Besancenot, J.-P., Thibaudon, M., 2016. Airborne Cladosporium fungal spores andclimate change in France. Aerobiologia 32, 53–68.

Skjøth, C.A., Sommer, J., Frederiksen, L., Gosewinkel Karlson, U., 2012. Crop harvest inDenmark and Central Europe contributes to the local load of airborne Alternariaspore concentrations in Copenhagen. Atmos. Chem. Phys. 12, 11107–11123.

Stockmarr, A., Andreasen, V., Ostergård, H., 2007. Dispersal distances for airborne sporesbased on deposition rates and stochastic modeling. Phytopathology 97, 1325–1330.

Straatsma, G., Ayer, F., Egli, S., 2001. Species richness, abundance, and phenology of fungalfruit bodies over 21 years in a Swiss forest plot. Mycol. Res. 105, 515–523.

Tariq, S.M., Matthews, S.M., Stevens, M., Hakim, E.A., 1996. Sensitization to Alternaria andCladosporium by the age of 4 years. Clin. Exp. Allergy 26, 794–798.

Thomma, B.P., van Esse, H.P., Crous, P.W., de Wit, P.J., 2005. Cladosporium fulvum (syn.Passalora fulva), a highly specialized plant pathogen as a model for functional studieson plant pathogenic Mycosphaerellaceae. Mol. Plant Pathol. 6, 379–393.

Van Leuken, J.P.G., Swart, A.N., Havelaar, A.H., Vanpul, A., Vanderhoek, W., Heederik, D.,2016. Atmospheric dispersion modelling of bioaerosols that are pathogenic tohumans and livestock – a review to inform risk assessment studies. Microb. RiskAnal. 1, 19–39.

van Maanen, A., Xu, X.-M., 2003. Modelling plant disease epidemics. Eur. J. Plant Pathol.109, 669–682.